Accurate day-ahead load forecasting is essential for optimizing distribution network planning. As the load data available to distribution networks becomes increasingly multidimensional and extensive, efficiently leveraging this data for precise day-ahead load forecasting has become a key research focus. To address this, an end-to-end approach that integrates data preprocessing, data decomposition,and data forecasting is proposed. In the data preprocessing stage, the bisecting K-means（BKM）clustering technique is used to reduce data noise and categorize the data, while combining dynamic and static feature extraction to capture load characteristics. In the data decomposition stage, the variational mode decomposition（VMD）technique is applied to decompose the preprocessed data into frequency components with strong periodicity and randomness. Finally, in the data forecasting stage, a temporal convolutional network（TCN）is employed to predict each mode component, and the predictions are aggregated to produce the final day-ahead load forecast. Case studies demonstrate that the BKM-VMD-TCN method proposed achieves superior forecasting accuracy compared to three other load forecasting methods.